Predictive Heating Optimization Accelerator
Deploy-ready Databricks Asset Bundle that runs Model Predictive Control over building/district-heating substations to cut heating energy 5-15% while holding comfort. Per-site thermal models, a real unit-tested NumPy MPC solver (no scipy/cvxpy), verified energy/cost/CO2 savings, and comfort-compliance monitoring. Fully generic and parameterized.
Product Preview
Try the interactive reader and demo tools below, or get the full product with all content unlocked.
⚙ Try Demo ToolsFile Structure 21 files
Documentation Preview README excerpt
Predictive Heating Optimization Accelerator
A deploy-ready Databricks Asset Bundle (DAB) that runs **Model Predictive
Control (MPC)** over district-heating substations — or any building portfolio
with supply-temperature control — to cut heating energy 5–15% while keeping
every building inside its comfort band.
Everything is generic and environment-agnostic: you point the bundle at your
own catalog, schema, warehouse, and source tables via bundle variables and
deploy with one command. There are no organization-specific catalogs, table
names, timezones, currencies, or business logic baked in.
What makes this different
- A real, working MPC solver —
src/optimization/mpc_solver.pyis a
complete, dependency-light optimizer in pure NumPy (no scipy/cvxpy). It is
unit-tested (tests/test_mpc_solver.py, 8 passing tests) and runs
offline. This is not a "implement your own solver here" placeholder.
- Verified savings, not estimated — savings are computed against a
reference baseline, with a comfort-compliance guardrail proving comfort was
never sacrificed for savings.
- One-command deploy — a standard DAB with three scheduled jobs.
How it works
1. Telemetry ingestion — substation temperatures, setpoints, weather.
2. Thermal model training (daily) — a per-site first-order linear model:
T_in(t+1) = alphaT_in(t) + betau(t) + gammaT_out(t) + deltarad(t) + c0
3. MPC optimization (every 15 min) — the solver computes the cost-optimal
supply-temperature trajectory against a price/marginal-cost signal, subject
to comfort and hardware bounds.
4. Setpoint dispatch — read optimization_results from your BMS/SCADA
integration and apply the first setpoint of each site's trajectory.
5. Savings & emissions (daily) — verified energy, cost, and CO2 savings.
Quickstart (30 minutes)
# 1. Install the Databricks CLI (>= 0.205.0)
curl -fsSL https://raw.githubusercontent.com/databricks/setup-cli/main/install.sh | sh
# 2. Configure a profile in ~/.databrickscfg (host + auth)
# 3. Set your variables in databricks.yml (catalog, schema, warehouse_id,
# telemetry_source, weather_source, price_source, timezone)
# 4. Validate and deploy
databricks bundle validate -t dev -p <your-profile>
databricks bundle deploy -t dev -p <your-profile>
Run the solver's test suite locally (no Databricks needed):
pip install numpy pytest
pytest tests/test_mpc_solver.py -v
Required data sources
... continues with setup instructions, usage examples, and more.